Efficient load frequency controller for a power system comprising renewable resources based on deep reinforcement learning

This paper presents the development of an adaptive load frequency controller (LFC) to mitigate frequency deviations in power systems comprising renewable energy sources (RESs) during transient and steady-state conditions. Integrating RESs with power systems results in frequency problems due to reduced system inertia and the intermittency of the RESs. The paper introduces a model-free controller that employs deep neural networks trained by the twin-delayed deep deterministic gradient reinforcement learning policy to generate the load reference signal (LRS) for the speed governor. The LRS is produced by the controller's agent, which undergoes training by receiving observations and rewards from the power system. These observations capture frequency errors resulting from load disturbances and renewable power fluctuations, while the reward assesses the controller's effectiveness in minimizing frequency errors. Compared to heuristic-based controllers, the proposed controller demonstrates considerable improvements in frequency stability for both steady-state error and transient response across various load disturbances when compared to heuristic-based controllers. Moreover, the proposed controller could limit the frequency deviations under varying weather conditions.